Post on 25-Aug-2018
1
A Description Logic with Concrete Domains
CS848 presentationPresenter: Yongjuan Zou
2
Reference
“A scheme for integrating concrete domains into concept languages” by Franz Baader and PhilippHanschke“A description logic with concrete domains and a role-forming predicate operator” by Volker Haarslev
“Visual spatial query languages: a semantics using description logic” by Volker Haarslev, Ralf MÖller and Michael Wessel
3
Outline
Description logic ALC(D)Extended language ALCRP(D)Combination of formal conceptual and spatial reasoning (Semantics of spatial queries)Discussions
4
The Description Logic ALC(D)Motivation
pure KL-ONE language?abstract logical level
ALC ?not allow for built-in predicates
Concrete domainR : set of all real numbers
S2 : set of all two-dimensional polygons
TIA: set of all time intervals
5
The Description Logic ALC(D)Motivation
Abstract objects can be related to concrete objects via features (functional roles)Relationships between concrete objects are described with a set of domain-specific predicatesConcrete domain provides access to domain-specific reasoning algorithmsMain idea: properties of abstract objects can be expressed using a concept-forming predicate operator.
6
The Description Logic ALC(D) Terminologies
Concrete DomainsA concrete domain: a pair of (∆D, ФD)n-ary predicate PD ⊆ ∆D
n
Admissible iffФD is closed under negation, contains a name TD for ∆Dthe satisfiability problem for finite conjunctions of predicates is decidable
7
The Description Logic ALC(D) Terminologies
Concept Terms C, R, and F : disjoint sets of concepts, role, and feature( i.e., function role) namesAny element of C is an atomic concept termLet C and D be concept terms, let R be an arbitrary role term ora feature, P∈ФD is a predicate name with arity n, and u1,...,un be feature chains (i.e., a composition of features), then the following expressions are also concept terms:C ⊔ D (disjunction),
C ⊓ D (conjunction),¬C (negation),∃R.C (concept exists restriction),∀R.C (concept value restriction), ∃u1,...,un.P (predicate exists restriction)
8
The Description Logic ALC(D) Terminologies
TerminologyA: a concept nameD: a concept termterminological axioms: A ≐ D, A ⊑ D
T is a terminology or TBox if no concept name in Tappears more than once on the left-hand side of a definition and, furthermore, if no cyclic definitions are present.
9
The Description Logic ALC(D) Terminologies
Semantics
10
The Description Logic ALC(D) The Assertional Language
SyntaxLet IA and ID be two disjoint sets of individual names for the abstract and concrete domain. If C is a concept term, R be a role name, f be a feature name, P be an n-ary predicate name of D, and a and b be elements of IA and x, x1,...,xn be elements of ID, then the following are assertional axioms:
a: C (concept membership), (a, b): R (role filler), (a, x): f (feature filler), (x1,...,xn) : P (predicate membership)SemanticsAn interpretation for the assertional language is an interpretation for the concept language which additionally maps every individual name from IA to a single element of ∆I and every individual name from ID to a single element of ∆D. Such an interpretation satisfies an assertional axiom
DIn
In
III
IIIII
PxxiffPxxxafifffxaRbaiffRbaCaiffCa
∈=
∈∈
),...(:),...,(,)(:),(),(:),(,:
11
11
The Description Logic ALC(D)Inference
Theorem: Let D be an admissible concrete domain. Then there exists a sound and complete algorithm which is able to decide the consistency of an ABoxfor ALC(D). ( Franz Baader et. al.)
Reasoning with ALC(D) is PSPACE-complete.
12
The Description Logic ALCRP(D)Motivation
Inferences about qualitative relations should not be considered in isolation but should be integrated with formal inferences about structural descriptions of domain objects and inferences about quantitative data. (inferences about spatial relations in GIS)
extends ALC(D) by introducing defined roles that are based on a role-forming predicate operator
13
The Description Logic ALCRP(D)Motivation
14
The Description Logic ALCRP(D)Terminologies
...Role Terms
15
The Description Logic ALCRP(D)Terminologies
ExampleAdmissible concrete domain RS2: union of R: all real numbers with predicates built by first order means from (in)equalities between integer polynomials in several indeterminates
S2: all two-dimensional polygons with topological relations as predicates
16
The Description Logic ALCRP(D)Terminologies
Exampledescription :“ a cottage that is enclosed by a forest”
get the spatial area via the feature has_area
topological predicate in S2
17
The Description Logic ALCRP(D)Terminologies
TerminologyA: a concept nameD: a concept termterminological axioms: A ≐ D, A ⊑ D
T is a terminology or TBox if no concept or rolename in T appears more than once on the left-hand side of a definition and, furthermore, if no cyclic definitions are present.
18
The Description Logic ALCRP(D)Terminologies
Semantics...
Example
19
The Description Logic ALCRP(D) The Assertional Language
Syntax and Semantics...R : an atomic or complex role term...
20
The Description Logic ALCRP(D) Inference
Theorem: The problem whether an ALCRP(D)concept term C is satisfiable w.r.t. a TBox Tis undecidable. ( Volker Haarslev, et,al.)
Two optionsRestrict the structure of the concrete domain predicatesRestrict the ability to combine some critical operators
21
The Description Logic ALCRP(D) Inference
22
The Description Logic ALCRP(D) Inference
C, D : concept namesRa : an atomic role termRc : a complex role termf : a featureu : a feature chain with a
length greater than 1
Theorem: The ABox consistency problem for restricted ALCRP(D)ABoxes, subsumption problem and satisfiability problem for ALCRP(D)concept terms are decidable w.r.t. terminologies for which the considered concept terms are restricted. ( Volker Haarslev)
23
Semantics of Spatial QueriesA VISCO application
MotivationThe specification of queries in a GIS could be made easier by combining spatial and terminological reasoning with visual language theory.
24
Semantics of Spatial QueriesA VISCO application
Visual spatial query: user draw a constellation of spatial entities that resemble the intended constellation of interest, with annotations of concept namesParser: analyze the drawing and create a corresponding ABox as semantic representationA Completion facility: resolve semantic ambiguities or to complete underspecified information by using default rules for further specialization
25
Semantics of Spatial QueriesA VISCO application
Completion of queriesDefault knowledge, if applied in a consistent way, can make queries preciseThe process of formulating queries be facilitated by automatically completing queries in a meaningful way
26
Semantics of Spatial QueriesA VISCO application
different possible worlds
Formal derivation processes
computing plausible candidates
ts
Conceptual information
Spatial relations
between domain objec
27
Semantics of Spatial QueriesA VISCO application
Reasoning about Visual Spatial QueriesExampleA buyer want a cottage:
28
Semantics of Spatial QueriesA VISCO application
Abstraction process: reduce particular ABoxesto corresponding ABoxes consisting of a single concept assertion representing the original query--- semantics of this query, say, a query concept, used to retrieve all “matching” individuals and answer the query.
29
Semantics of Spatial QueriesA VISCO application
}cottage:{1c1 cA =
cottagec1 is classified by the reasoner.
The semantic validity of the query is automatically verified during classification.
If the forest f were required to be ‘mosquito-free’, the ALCRP(D)reasoner would recognize the incoherence of cottagec1.and identify the source of contradiction.
30
Semantics of Spatial QueriesA VISCO application
Query optimizationsubsumption between query concepts
Query optimizer: detect query subsumption and reduce the search space to the set of query matches already computed
31
Semantics of Spatial QueriesA VISCO application
Contributionthe first proposal utilizing an expressive and decidable spatial logic to specify the semantics of visual spatial queries
specification of a semanticsreasoning about query subsumptionreasoning about applying default knowledge
32
Discussions
Representing Spatiotemporal PhenomenaTIA : a concrete domain representing time intervals with two-place predicates representing Allen’s Interval Algebra
Description logic ALCRP(S2 ‘TIA)...any two disjoint admissible concrete domains can be combined to form a
single admissible concrete domain.(Ref.[2])
example application: city construction planning...
⊕
33
Discussions
Formalism built on the clean integration of Description Logics and concrete domains.Open problem: if better results with respect to decidability can be obtained by designing a special-purpose (e.g. topological) description logic?
The general question is under which conditions a special-purpose description logic can provide more expressive power than a generic one while still remaining decidable.